러프집합을 이용한 감별진단에 필요한 주요 항목 선별

Abstract

응급실에 내원하는 환자들 중 가장 흔한 증상 중에 하나가 복통이며, 다양한 원인과 증상을 보이기 때문에 감별진단하기가 쉽지 않다. 또한 신속한 처치가 지연될 경우 환자의 예후에 심각한 영향을 미칠 수 있다. 따라서 최소의 검사만으로 신뢰할 수 있는 결과를 제공할 수 있는 다양한 특징 추출에 관한 연구가 제안되었다. 대표적으로 카이제곱검정, 클러스터링, 퍼지집합 등 이 있다. 그러나 임상 데이터의 복잡성과 불명확성 때문에 최적의 결과를 획득하는 것에는 제한점이 있다. 따라서 본 논문에서는 러프집합의 식별가능행렬 및 식별가능함수를 이용하여 주증상을 포함한 임상검사항목(즉, 혈액검사 및 요검사)에서 중요 항목을 선택하고, 감별진단을 위한 지식을 획득할 수 있는 방법을 제안하였다. 실험에서는 제안된 방법의 효과성을 보이기 위해서 2006년 7월부터 2007년 6월까지 응급실에 복통을 주증상으로 내원한 1,103명의 환자들을 대상으로 주증상을 포함한 52가지 검사항목의 결과를 근거로 3가지 다빈도 질환(급성 충수염, 담낭염 및 담석증)의 감별진단 시 중요 검사항목을 추출하고 지식을 생성하였다. 또한 전체 1,103명의 환자들의 입원과 퇴원을 판별하기 위해 주요한 검사항목과 지식을 획득하였으며, 카이제곱검정(p<0.05) 결과와 비교하였다. 실험결과, 급성 충수염과 담낭염·담석증을 감별 진단을 위한 주요 항목은 각각 (complaints, urine glucose, urinary ketone body, urine occult blood, urine white blood cell 1, urine red blood cell, white blood cell, hematocrit, mean corpuscular hemoglobin, neutrophil, monocyte, mean platelet volume, chloride, total amylase, glucose, total cholesterol, alkaline phosphatase, alanine aminotransferase}와 (complaints, urine albumin, urinary ketone body, urine occult blood, urine red blood cell, white blood cell, hematocrit, mean corpuscular hemoglobin, lymphocyte, monocyte, chloride, total amylase, glucose, direct bilirubin}이 추출되었고, 이들 두 질환을 감별 진단하는 주요 항목은 (complaints, urine glucose, urine occult blood, urine white blood cell 1, white blood cell, hematocrit, mean corpuscular hemoglobin, chloride, total amylase, albumin, alanine aminotransferase)이 추출되었다. 또한 1,103명의 환자들의 입원과 퇴원을 판별하는 주요 항목은 (complaints, urinary ketone body, urine occult blood, urine white blood cell 1, urine white blood cell 2, urine red blood cell, epithelial cell, hematocrit, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, lymphocyte, monocyte, chloride, total amylase, glucose, blood urea nittrogen, total cholesterol, total bilirubin, alkaline phosphatase, alanine aminotransferase}이 검출된 항목임을 확인할 수 있었다.

One of the most common symptoms among the patients admitted to the emergency room is a stomachache, which varies widely in causes and symptoms and accordingly is difficult to diagnose. When fast treatment is delayed, it can have serious impacts on the patient's prognosis. Thus suggested was a study on extracting diverse characteristics to provide reliable results only with a minimum set of tests. Good examples include the chi-square test, clustering, and fuzzy sets. Due to the complexity and obscurity of the clinical data, however, there are restrictions to the acquisition of optimal results. Thus this study introduced a method of selecting significance items from the clinical test list (i.e. a blood and urine test) including the major symptoms and acquiring knowledge needed for differential diagnoses by using the discernibility matrix and function of rough sets. In order to demonstrate the effectiveness of the proposed method, an experiment was conducted which extracted significance test items for making a differential diagnosis of the three most frequent diseases (acute appendicitis, cholecystitis, and cholelithiasis) based on the results of the 52 test items including the major symptoms reported by 1,103 patients, who were admitted to the emergency room due to a stomachache as the major system from July, 2006 to June, 2007, and generated a pool of knowledge. In addition, another set of significance test items and knowledge was obtained to distinguish admission and discharge among the 1,103 patients and compared with the results of the chi-square test (p<0.05). As a result, the major items for the differential diagnosis of acute appendicitis and cholecystitis·cholelithiasis were (complaints, urine glucose, urinary ketone body, urine occult blood, urine white blood cell 1, urine red blood cell, white blood cell, hematocrit, mean corpuscular hemoglobin, neutrophil, monocyte, mean platelet volume, chloride, total amylase, glucose, total cholesterol, alkaline phosphatase, alanine aminotransferase) and (complaints, urine albumin, urinary ketone body, urine occult blood, urine red blood cell, white blood cell, hematocrit, mean corpuscular hemoglobin, lymphocyte, monocyte, chloride, total amylase, glucose, direct bilirubin), respectively, those for the two diseases were (complaints, urine glucose, urine occult blood, urine white blood cell 1, white blood cell, hematocrit, mean corpuscular hemoglobin, chloride, total amylase, albumin, alanine aminotransferase), and those for distinguishing admission and discharge among the 1,103 patients were (complaints, urinary ketone body, urine occult blood, urine white blood cell 1, urine white blood cell 2, urine red blood cell, epithelial cell, hematocrit, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, lymphocyte, monocyte, chloride, total amylase, glucose, blood urea nittrogen, total cholesterol, total bilirubin, alkaline phosphatase, alanine aminotransferase).